def scrub(self):
"""
The XML file seems to have mixed-encoding;
we scrub out the control characters
from the file for processing.
i.e.?i
omia.xml:1555328.28: PCDATA invalid Char value 2
<field name="journal">Bulletin et Memoires de la Societe Centrale de Medic
:return:
"""
logger.info(
"Scrubbing out the nasty characters that break our parser.")
myfile = '/'.join((self.rawdir, self.files['data']['file']))
tmpfile = '/'.join((self.rawdir, self.files['data']['file']+'.tmp.gz'))
t = gzip.open(tmpfile, 'wb')
du = DipperUtil()
with gzip.open(myfile, 'rb') as f:
filereader = io.TextIOWrapper(f, newline="")
for l in filereader:
l = du.remove_control_characters(l) + '\n'
t.write(l.encode('utf-8'))
t.close()
# TEC I do not like this at all. original data must be preserved as is.
# also may be heavy handed as chars which do not break the parser
# are stripped as well (i.e. tabs and newlines)
# move the temp file
logger.info("Replacing the original data with the scrubbed file.")
shutil.move(tmpfile, myfile)
return
def _process_data(self, source, limit=None):
"""
This function will process the data files from Coriell.
We make the assumption that any alleles listed are variants
(alternates to w.t.)
Triples: (examples)
:NIGMSrepository a CLO_0000008 #repository
label : NIGMS Human Genetic Cell Repository
foaf:page
https://catalog.coriell.org/0/sections/collections/NIGMS/?SsId=8
line_id a CL_0000057, #fibroblast line
derives_from patient_id
part_of :NIGMSrepository
RO:model_of OMIM:disease_id
patient id a foaf:person,
label: "fibroblast from patient 12345 with disease X"
member_of family_id #what is the right thing here?
SIO:race EFO:caucasian #subclass of EFO:0001799
in_taxon NCBITaxon:9606
dc:description Literal(remark)
RO:has_phenotype OMIM:disease_id
GENO:has_genotype genotype_id
family_id a owl:NamedIndividual
foaf:page
"https://catalog.coriell.org/0/Sections/BrowseCatalog/FamilyTypeSubDetail.aspx?PgId=402&fam=2104&coll=GM"
genotype_id a intrinsic_genotype
GENO:has_alternate_part allelic_variant_id
we don't necessarily know much about the genotype,
other than the allelic variant. also there's the sex here
pub_id mentions cell_line_id
:param raw:
:param limit:
:return:
"""
raw = '/'.join((self.rawdir, self.files[source]['file']))
LOG.info("Processing Data from %s", raw)
if self.testMode: # set the graph to build
graph = self.testgraph
else:
graph = self.graph
family = Family(graph)
model = Model(graph)
line_counter = 1
geno = Genotype(graph)
diputil = DipperUtil()
col = self.files[source]['columns']
# affords access with
# x = row[col.index('x')].strip()
with open(raw, 'r', encoding="iso-8859-1") as csvfile:
filereader = csv.reader(csvfile, delimiter=',', quotechar=r'"')
# we can keep a close watch on changing file formats
fileheader = next(filereader, None)
fileheader = [c.lower() for c in fileheader]
if col != fileheader: # assert
LOG.error('Expected %s to have columns: %s', raw, col)
LOG.error('But Found %s to have columns: %s', raw, fileheader)
raise AssertionError('Incomming data headers have changed.')
for row in filereader:
line_counter += 1
if len(row) != len(col):
LOG.warning('Expected %i values but find %i in row %i',
len(col), len(row), line_counter)
continue
# (catalog_id, description, omim_number, sample_type,
# cell_line_available, dna_in_stock, dna_ref, gender, age,
# race, ethnicity, affected, karyotype, relprob, mutation,
# gene, family_id, collection, url, cat_remark, pubmed_ids,
# family_member, variant_id, dbsnp_id, species) = row
# example:
# GM00003,HURLER SYNDROME,607014,Fibroblast,Yes,No,
# ,Female,26 YR,Caucasian,,,,
# parent,,,39,NIGMS Human Genetic Cell Repository,
# http://ccr.coriell.org/Sections/Search/Sample_Detail.aspx?Ref=GM00003,
# 46;XX; clinically normal mother of a child with Hurler syndrome;
# proband not in Repository,,
# 2,,18343,H**o sapiens
catalog_id = row[col.index('catalog_id')].strip()
if self.testMode and catalog_id not in self.test_lines:
# skip rows not in our test lines, when in test mode
continue
# ########### BUILD REQUIRED VARIABLES ###########
# Make the cell line ID
cell_line_id = 'Coriell:' + catalog_id
# Map the cell/sample type
cell_type = self.resolve(row[col.index('sample_type')].strip())
# on fail cell_type = self.globaltt['cell'] ?
# Make a cell line label
collection = row[col.index('collection')].strip()
line_label = collection.partition(' ')[0] + '-' + catalog_id
# Map the repository/collection
repository = self.localtt[collection]
# patients are uniquely identified by one of:
# dbsnp id (which is == an individual haplotype)
# family id + family member (if present) OR
# probands are usually family member zero
# cell line id
# since some patients have >1 cell line derived from them,
# we must make sure that the genotype is attached to
# the patient, and can be inferred to the cell line
# examples of repeated patients are:
# famid=1159, member=1; fam=152,member=1
# Make the patient ID
# make an anonymous patient
patient_id = '_:person'
fam_id = row[col.index('fam')].strip()
fammember = row[col.index('fammember')].strip()
if fam_id != '':
patient_id = '-'.join((patient_id, fam_id, fammember))
else:
# make an anonymous patient
patient_id = '-'.join((patient_id, catalog_id))
# properties of the individual patients: sex, family id,
# member/relproband, description descriptions are
# really long and ugly SCREAMING text, so need to clean up
# the control cases are so odd with this labeling scheme;
# but we'll deal with it as-is for now.
description = row[col.index('description')].strip()
short_desc = (description.split(';')[0]).capitalize()
gender = row[col.index('gender')].strip().lower()
affected = row[col.index('affected')].strip()
relprob = row[col.index('relprob')].strip()
if affected == '':
affected = 'unspecified'
elif affected in self.localtt:
affected = self.localtt[affected]
else:
LOG.warning('Novel Affected status %s at row: %i of %s',
affected, line_counter, raw)
patient_label = ' '.join((affected, gender, relprob))
if relprob == 'proband':
patient_label = ' '.join(
(patient_label.strip(), 'with', short_desc))
else:
patient_label = ' '.join(
(patient_label.strip(), 'of proband with', short_desc))
# ############# BUILD THE CELL LINE #############
# Adding the cell line as a typed individual.
cell_line_reagent_id = self.globaltt['cell line']
model.addIndividualToGraph(cell_line_id, line_label,
cell_line_reagent_id)
# add the equivalent id == dna_ref
dna_ref = row[col.index('dna_ref')].strip()
if dna_ref != '' and dna_ref != catalog_id:
equiv_cell_line = 'Coriell:' + dna_ref
# some of the equivalent ids are not defined
# in the source data; so add them
model.addIndividualToGraph(equiv_cell_line, None,
cell_line_reagent_id)
model.addSameIndividual(cell_line_id, equiv_cell_line)
# Cell line derives from patient
geno.addDerivesFrom(cell_line_id, patient_id)
geno.addDerivesFrom(cell_line_id, cell_type)
# Cell line a member of repository
family.addMember(repository, cell_line_id)
cat_remark = row[col.index('cat_remark')].strip()
if cat_remark != '':
model.addDescription(cell_line_id, cat_remark)
# Cell age_at_sampling
# TODO add the age nodes when modeled properly in #78
# if (age != ''):
# this would give a BNode that is an instance of Age.
# but i don't know how to connect
# the age node to the cell line? we need to ask @mbrush
# age_id = '_'+re.sub('\s+','_',age)
# gu.addIndividualToGraph(
# graph,age_id,age,self.globaltt['age'])
# gu.addTriple(
# graph,age_id,self.globaltt['has measurement value'],age,
# True)
# ############# BUILD THE PATIENT #############
# Add the patient ID as an individual.
model.addPerson(patient_id, patient_label)
# TODO map relationship to proband as a class
# (what ontology?)
# Add race of patient
# FIXME: Adjust for subcategories based on ethnicity field
# EDIT: There are 743 different entries for ethnicity...
# Too many to map?
# Add ethnicity as literal in addition to the mapped race?
# Adjust the ethnicity txt (if using)
# to initial capitalization to remove ALLCAPS
# TODO race should go into the individual's background
# and abstracted out to the Genotype class punting for now.
# if race != '':
# mapped_race = self.resolve(race)
# if mapped_race is not None:
# gu.addTriple(
# g,patient_id,self.globaltt['race'], mapped_race)
# model.addSubClass(
# mapped_race,self.globaltt['ethnic_group'])
# ############# BUILD THE FAMILY #############
# Add triples for family_id, if present.
if fam_id != '':
family_comp_id = 'CoriellFamily:' + fam_id
family_label = ' '.join(
('Family of proband with', short_desc))
# Add the family ID as a named individual
model.addIndividualToGraph(family_comp_id, family_label,
self.globaltt['family'])
# Add the patient as a member of the family
family.addMemberOf(patient_id, family_comp_id)
# ############# BUILD THE GENOTYPE #############
# the important things to pay attention to here are:
# karyotype = chr rearrangements (somatic?)
# mutation = protein-level mutation as a label,
# often from omim
# gene = gene symbol - TODO get id
# variant_id = omim variant ids (; delimited)
# dbsnp_id = snp individual ids = full genotype?
# note GM00633 is a good example of chromosomal variation
# - do we have enough to capture this?
# GM00325 has both abnormal karyotype and variation
# make an assumption that if the taxon is blank,
# that it is human!
species = row[col.index('species')].strip()
if species is None or species == '':
species = 'H**o sapiens'
taxon = self.resolve(species)
# if there's a dbSNP id,
# this is actually the individual's genotype
genotype_id = None
genotype_label = None
dbsnp_id = row[col.index('dbsnp_id')].strip()
if dbsnp_id != '':
genotype_id = 'dbSNPIndividual:' + dbsnp_id
omim_map = {}
gvc_id = None
# some of the karyotypes are encoded
# with terrible hidden codes. remove them here
# i've seen a <98> character
karyotype = row[col.index('karyotype')].strip()
karyotype = diputil.remove_control_characters(karyotype)
karyotype_id = None
if karyotype.strip() != '':
karyotype_id = '_:' + re.sub('MONARCH:', '',
self.make_id(karyotype))
# add karyotype as karyotype_variation_complement
model.addIndividualToGraph(
karyotype_id, karyotype,
self.globaltt['karyotype_variation_complement'])
# TODO break down the karyotype into parts
# and map into GENO. depends on #77
# place the karyotype in a location(s).
karyo_chrs = self._get_affected_chromosomes_from_karyotype(
karyotype)
for chrom in karyo_chrs:
chr_id = makeChromID(chrom, taxon, 'CHR')
# add an anonymous sequence feature,
# each located on chr
karyotype_feature_id = '-'.join((karyotype_id, chrom))
karyotype_feature_label = \
'some karyotype alteration on chr' + str(chrom)
feat = Feature(graph, karyotype_feature_id,
karyotype_feature_label,
self.globaltt['sequence_alteration'])
feat.addFeatureStartLocation(None, chr_id)
feat.addFeatureToGraph()
geno.addParts(karyotype_feature_id, karyotype_id,
self.globaltt['has_variant_part'])
gene = row[col.index('gene')].strip()
mutation = row[col.index('mutation')].strip()
if gene != '':
vl = gene + '(' + mutation + ')'
# fix the variant_id so it's always in the same order
variant_id = row[col.index('variant_id')].strip()
vids = variant_id.split(';')
variant_id = ';'.join(sorted(list(set(vids))))
if karyotype.strip() != '' and not self._is_normal_karyotype(
karyotype):
gvc_id = karyotype_id
if variant_id != '':
gvc_id = '_:' + variant_id.replace(';', '-') + '-' \
+ re.sub(r'\w*:', '', karyotype_id)
if mutation.strip() != '':
gvc_label = '; '.join((vl, karyotype))
else:
gvc_label = karyotype
elif variant_id.strip() != '':
gvc_id = '_:' + variant_id.replace(';', '-')
gvc_label = vl
else:
# wildtype?
pass
# add the karyotype to the gvc.
# use reference if normal karyotype
karyo_rel = self.globaltt['has_variant_part']
if self._is_normal_karyotype(karyotype):
karyo_rel = self.globaltt['has_reference_part']
if karyotype_id is not None \
and not self._is_normal_karyotype(karyotype) \
and gvc_id is not None and karyotype_id != gvc_id:
geno.addParts(karyotype_id, gvc_id, karyo_rel)
if variant_id.strip() != '':
# split the variants & add them as part of the genotype
# we don't necessarily know their zygosity,
# just that they are part of the genotype variant ids
# are from OMIM, so prefix as such we assume that the
# sequence alts will be defined in OMIM not here
# TODO sort the variant_id list, if the omim prefix is
# the same, then assume it's the locus make a hashmap
# of the omim id to variant id list;
# then build the genotype hashmap is also useful for
# removing the "genes" from the list of "phenotypes"
# will hold gene/locus id to variant list
omim_map = {}
locus_num = None
for var in variant_id.split(';'):
# handle omim-style and odd var ids
# like 610661.p.R401X
mch = re.match(r'(\d+)\.+(.*)', var.strip())
if mch is not None and len(mch.groups()) == 2:
(locus_num, var_num) = mch.groups()
if locus_num is not None and locus_num not in omim_map:
omim_map[locus_num] = [var_num]
else:
omim_map[locus_num] += [var_num]
for omim in omim_map:
# gene_id = 'OMIM:' + omim # TODO unused
vslc_id = '_:' + '-'.join(
[omim + '.' + a for a in omim_map.get(omim)])
vslc_label = vl
# we don't really know the zygosity of
# the alleles at all.
# so the vslcs are just a pot of them
model.addIndividualToGraph(
vslc_id, vslc_label,
self.globaltt['variant single locus complement'])
for var in omim_map.get(omim):
# this is actually a sequence alt
allele1_id = 'OMIM:' + omim + '.' + var
geno.addSequenceAlteration(allele1_id, None)
# assume that the sa -> var_loc -> gene
# is taken care of in OMIM
geno.addPartsToVSLC(
vslc_id, allele1_id, None,
self.globaltt['indeterminate'],
self.globaltt['has_variant_part'])
if vslc_id != gvc_id:
geno.addVSLCtoParent(vslc_id, gvc_id)
if affected == 'unaffected':
# let's just say that this person is wildtype
model.addType(patient_id, self.globaltt['wildtype'])
elif genotype_id is None:
# make an anonymous genotype id (aka blank node)
genotype_id = '_:geno' + catalog_id.strip()
# add the gvc
if gvc_id is not None:
model.addIndividualToGraph(
gvc_id, gvc_label,
self.globaltt['genomic_variation_complement'])
# add the gvc to the genotype
if genotype_id is not None:
if affected == 'unaffected':
rel = self.globaltt['has_reference_part']
else:
rel = self.globaltt['has_variant_part']
geno.addParts(gvc_id, genotype_id, rel)
if karyotype_id is not None \
and self._is_normal_karyotype(karyotype):
if gvc_label is not None and gvc_label != '':
genotype_label = '; '.join((gvc_label, karyotype))
elif karyotype is not None:
genotype_label = karyotype
if genotype_id is None:
genotype_id = karyotype_id
else:
geno.addParts(karyotype_id, genotype_id,
self.globaltt['has_reference_part'])
else:
genotype_label = gvc_label
# use the catalog id as the background
genotype_label += ' [' + catalog_id.strip() + ']'
if genotype_id is not None and gvc_id is not None:
# only add the genotype if it has some parts
geno.addGenotype(genotype_id, genotype_label,
self.globaltt['intrinsic_genotype'])
geno.addTaxon(taxon, genotype_id)
# add that the patient has the genotype
# TODO check if the genotype belongs to
# the cell line or to the patient
graph.addTriple(patient_id, self.globaltt['has_genotype'],
genotype_id)
else:
geno.addTaxon(taxon, patient_id)
# TODO: Add sex/gender (as part of the karyotype?)
# = row[col.index('')].strip()
# ############# DEAL WITH THE DISEASES #############
omim_num = row[col.index('omim_num')].strip()
# we associate the disease to the patient
if affected == 'affected' and omim_num != '':
for d in omim_num.split(';'):
if d is not None and d != '':
# if the omim number is in omim_map,
# then it is a gene not a pheno
# TEC - another place to use the mimTitle omim
# classifier omia & genereviews are using
if d not in omim_map:
disease_id = 'OMIM:' + d.strip()
# assume the label is taken care of in OMIM
model.addClassToGraph(disease_id, None)
# add the association:
# the patient has the disease
assoc = G2PAssoc(graph, self.name, patient_id,
disease_id)
assoc.add_association_to_graph()
# this line is a model of this disease
# TODO abstract out model into
# it's own association class?
graph.addTriple(cell_line_id,
self.globaltt['is model of'],
disease_id)
else:
LOG.info('drop gene %s from disease list', d)
# ############# ADD PUBLICATIONS #############
pubmed_ids = row[col.index('pubmed_ids')].strip()
if pubmed_ids != '':
for s in pubmed_ids.split(';'):
pubmed_id = 'PMID:' + s.strip()
ref = Reference(graph, pubmed_id)
ref.setType(self.globaltt['journal article'])
ref.addRefToGraph()
graph.addTriple(pubmed_id, self.globaltt['mentions'],
cell_line_id)
if not self.testMode and (limit is not None
and line_counter > limit):
break
return
def _add_variant_gene_relationship(self, patient_var_map,
gene_coordinate_map):
"""
Right now it is unclear the best approach on how to connect
variants to genes. In most cases has_affected_locus/GENO:0000418
is accurate; however, there are cases where a variant is in the intron
on one gene and is purported to causally affect another gene down or
upstream. In these cases we must first disambiguate which gene
is the affected locus, and which gene(s) are predicated to be
causully influenced by (RO:0002566)
UPDATE 8-30: In the latest dataset we no longer have 1-many mappings
between variants and genes, but leaving this here in case we see
these in the future
The logic followed here is:
if mutation type contains downstream/upstream and more than one
gene of interest, investigate coordinates of all genes to
see if we can disambiguate which genes are which
:return: None
"""
# genotype = Genotype(self.graph)
dipper_util = DipperUtil()
model = Model(self.graph)
# Note this could be compressed in someway to remove one level of for looping
for patient in patient_var_map:
for variant_id, variant in patient_var_map[patient].items():
variant_bnode = self.make_id("{0}".format(variant_id), "_")
genes_of_interest = variant['genes_of_interest']
if len(genes_of_interest) == 1:
# Assume variant is variant allele of gene
gene = genes_of_interest[0]
gene_id = dipper_util.get_hgnc_id_from_symbol(gene)
self._add_gene_to_graph(
gene, variant_bnode, gene_id,
self.globaltt['has_affected_feature'])
elif re.search(r'upstream|downstream',
variant['type'],
flags=re.I):
# Attempt to disambiguate
ref_gene = []
up_down_gene = []
unmatched_genes = []
for gene in variant['genes_of_interest']:
if gene_id and gene_id != '' and gene_id in gene_coordinate_map:
if gene_coordinate_map[gene_id]['start'] \
<= variant['position']\
<= gene_coordinate_map[gene_id]['end']:
gene_info = {
'symbol':
gene,
'strand':
gene_coordinate_map[gene_id]['strand']
}
ref_gene.append(gene_info)
else:
up_down_gene.append(gene)
else:
unmatched_genes.append(gene)
if len(ref_gene) == 1:
self._add_gene_to_graph(
ref_gene[0]['symbol'], variant_bnode, gene_id,
self.globaltt['has_affected_feature'])
# update label with gene
gene_list = [ref_gene[0]['symbol']
] # build label expects list
variant_label = self._build_variant_label(
variant['build'], variant['chromosome'],
variant['position'], variant['reference_allele'],
variant['variant_allele'], gene_list)
model.addLabel(variant_bnode, variant_label)
# In some cases there are multiple instances
# of same gene from dupe rows in the source
# Credit http://stackoverflow.com/a/3844832
elif len(ref_gene) > 0 and ref_gene[1:] == ref_gene[:-1]:
self._add_gene_to_graph(
ref_gene[0]['symbol'], variant_bnode, gene_id,
self.globaltt['has_affected_feature'])
# build label function expects list
gene_list = [ref_gene[0]['symbol']]
variant_label = self._build_variant_label(
variant['build'], variant['chromosome'],
variant['position'], variant['reference_allele'],
variant['variant_allele'], gene_list)
model.addLabel(variant_bnode, variant_label)
# Check if reference genes are on different strands
elif len(ref_gene) == 2:
strands = [st['strand'] for st in ref_gene]
if "minus" in strands and "plus" in strands:
for r_gene in ref_gene:
self._add_gene_to_graph(
r_gene['symbol'], variant_bnode, gene_id,
self.globaltt['has_affected_feature'])
else:
LOG.warning(
"unable to map intron variant to gene coordinates: %s",
variant)
for r_gene in ref_gene:
self._add_gene_to_graph(
r_gene['symbol'], variant_bnode, gene_id,
self.globaltt['causally_influences'])
elif re.search(r'intron', variant['type'], flags=re.I):
LOG.warning(
"unable to map intron variant to gene coordinates_2: %s",
variant)
for neighbor in up_down_gene:
self._add_gene_to_graph(
neighbor, variant_bnode, gene_id,
self.globaltt['causally_influences'])
# Unmatched genes are likely because we cannot map to an NCBIGene
# or we do not have coordinate information
for unmatched_gene in unmatched_genes:
self._add_gene_to_graph(
unmatched_gene, variant_bnode, gene_id,
self.globaltt['causally_influences'])
return
